First Years Away From Home: Letting Go and Staying Connected

This is a randomized controlled trial (RCT) with two experimental conditions (Handbook and
Handbook+) and a control. Each condition will be composed of 450 student/parent dyads, about
half in Cohort 1 and half in Cohort 2. Participants will be recruited from the incoming
freshman classes at Washington State University in 2017 and 2018 by the Social Development
Research Group (SDRG) at University of Washington. Parents in both experimental conditions
will receive a handbook in the summer before school starts. The handbook contains suggested
conversation starters, checklists, and activities to be completed by the parent/student dyad.
Activities are based in theory of social and emotional development and designed to address
risk and protective factors for this age group as they transition to living independently
from parents. Handbook plus parents will additionally receive text messages periodically
throughout the fall and spring semesters, some of which are designed to let them know about
major university events (e.g. home football games) which are known to be high-risk periods.
Each cohort will be followed for two years. Parents complete surveys at baseline (spring
before students starts) and in fall semester. Students complete surveys at baseline and each
semester for their first two years at school.

Specific Aims of the project are as follows:

Aim 1: Test the efficacy of Letting Go and Staying Connected, a theoretically guided,
developmentally targeted, and empirically supported handbook intervention for parents of
students transitioning to college. The investigators hypothesize that students in the
Handbook conditions will engage in fewer risk behaviors, including initiation, use, and abuse
of substances (alcohol, marijuana, opiates, and prescription stimulants) and HIV/sex risk
behaviors. As a result, Handbook students will also report fewer physical, emotional, social,
and academic harms. The investigators will test a control condition vs. a "Handbook Only" and
a "Handbook Plus" condition (an enhanced version of the intervention that adds
strategically-timed Short Message Service and/or emails designed to increase utilization of
the intervention).

Aim 2: Test theoretically specified mediational pathways between Handbook utilization and the
expected intervention changes in risk behaviors/associated harms. The investigators expect
that the effect of the Handbook on student outcomes will be mediated by both parent and
student behaviors, as described in the following path: 1) Parent engagement in and use of the
intervention 2) Parent practices in family management, autonomy support, and emotional
support 3) Students' perceived rewards for healthy behaviors 4) Healthy connections with
parents and peers 5) Healthy beliefs and clear standards leading to 6) Reduced substance use
and sex risk behaviors; increased academic performance.

Aim 3: Identify specific attributes of parent-student communication that predict and covary
with student substance use and HIV/sex risk behaviors. Early in their first semester, the
investigators will survey students intensively over 2 periods of 7 days to learn about their
daily communication with parents and their daily risk behaviors. Because little is known
about parent-student communication at this developmental stage, the investigators will
conduct descriptive and exploratory analyses on the content, emotional tone, timing,
duration, mode, and frequency of parents' communication with their students and examine how
those aspects of communication relate to student risk behaviors and associated harms at both
a between-person and a within-person level.

The SDRG has extensive experience in recruitment, data collection, data management, and data
security. The investigators' detailed Data Safety Monitoring Plan contains information about
specific procedures used to safeguard participant information. The investigators will recruit
underrepresented students in proportion to Washington State University (WSU) student
percentages. SDRG also has extensive experience in retaining study participants, in some
studies for more than 20 years.

Data cleaning. The investigators will conduct tests for out-of-range values, consistency
across measurement points, normality, and influential outliers. The investigators will
address any issues of bias related to outliers using accepted techniques (e.g. analyses
robust to violation of assumptions). Counts of dichotomous substance use variables are
frequently skewed; the investigators will use log transformations of variables and
appropriate analyses (e.g. survival analysis or Poisson regressions) as needed.

Selection and Attrition. Randomized selection of students for invitation into the study
should ensure a representative sample of the student population, and numerous studies have
demonstrated the ability of the Survey Research Division at the SDRG to retain over 90% of
respondents over more than 10 years. However, the investigators will test for differential
participation and attrition and for the likelihood of biased outcome estimates; if necessary,
the investigators will model those effects and employ correction techniques.

Variability in Implementation and Dosage. Particularly in the case of a self-administered
intervention, implementation and dosage are not distributed randomly -- for example, parents
in the intervention condition who are particularly motivated or compliant may do all
activities, whereas others may pick and choose. The non-random distribution of intervention
components across intervention participants confounds the ability to draw causal inferences
about program effects and obscures the mechanisms of program effects by treating
implementation as a binary event. In addition to intent-to-treat analyses the investigators
will employ recent methodological advances that allow for estimation, modeling, and control
of selection effects in implementation through use of propensity scoring and inverse
probability weighting.

Missing Data. When respondents skip items or miss one or more data collection points, missing
data may introduce bias and increase the likelihood of Type I errors. The investigators'
primary analytic approaches use Full Information Maximum Likelihood (FIML)(38) to manage
missing data. In case of excessive missingness, the investigators will use multiple
imputation .

Family-wise Error. The investigators will use summary variables of risk behaviors/harms, but
the investigators will also conduct separate analyses for each risk outcome to provide a
finer-grained look at individual substances, increasing the likelihood of a Type 1 error.
Responses to substance use items are also likely to be highly correlated; in such cases
standard Bonferroni corrections tend to be overly conservative with a higher likelihood of
Type 2 error. Resampling procedures (e.g. Statistical Analysis Software (SAS) Procedure
MULTTEST) are more powerful than the Bonferroni test and can be applied to categorical,
continuous, and censored data tests, mixed method analyses of between-subjects effects and
longitudinal within-subject tests(15,148). A global test statistic will be used when
appropriate.

Analyses: Aim 1: To measure intervention impact on change across time the investigators will
model latent growth curves (LGC) for each of the hypothesized outcomes. First, unconditional
models will be estimated to confirm the expected escalation across 5 waves of data. The
investigators expect a two factor model: an intercept factor set to pre-test levels, and a
linear slope factor estimated with factor loadings set to indicate pre-test as intercept and
non-equal intervals between surveys. A non-linear slope will be estimated using quadratic
forms for loadings, however the investigators expect primarily linearly increases over all 5
time points. A covariance is typically estimated between the intercept and slope factors to
control for pre-test levels of the outcome. Control variables will be added in the next step.
Finally, intervention group assignment (in the form of 2 dummy variables) will be added as
predictors of the mean and variance of the slope factor (random assignment should preclude
group effects on the intercept set at pre-test). The investigators expect these parameters to
reflect significant positive slope means and variances in the unconditional model and the
minimally conditional model. In the final models, intervention dummy variables are
hypothesized to have significant negative effects on slope mean and variance compared to the
control condition indicating both intervention conditions reduced the rate of escalation in
risk behaviors and consequences and also reduced the variability in these outcomes. By
manipulating the coding of dummy variables the investigators can then test the comparison
between H and H+ with the expectation that H+ parameters will be significantly more strongly
negative than those for H.

Aim 2: Preliminary intent-to-treat analyses (ANCOVA) will be conducted on mediators at each
follow up as described for outcomes above. Then LGC analyses described above will be extended
to include mediating variables. However, prior to estimating mediation, each mediator will be
modeled over time in a LGC model to determine its shape (intercept, linear slope, and
quadratic slope) conditioned on control variables as described for outcomes in Aim 1. In the
case of mediators the slope factors are less predictable, although in general these positive
parenting and parent-student relationship factors are expected to decline over time. Once
appropriate models of change have been determined, concurrent/parallel growth processes in
mediators and outcomes will be estimated. Parallel process Longitudinal Growth Modeling (LGM)
is a technique that has been used to examine interrelationships among various developmental
outcomes, including co-occurring drug use and delinquency, in adolescents and young adults
and for testing the mechanisms of preventive intervention effects. In the final step the
intervention group status dummy variables will be added to the parallel process models.

Aim 3: Given the exploratory nature of this aim, there are no hypotheses. Note that changes
in communication are not posited to mediate intervention effects. However, if significant
relationships are found between intervention status and communication measures, exploratory
analyses may be conducted on the control group only to better reflect etiological
relationships un-confounded by possible intervention influences. Preliminary analyses will
focus on describing levels and variability in content, emotional tone, timing, duration,
mode, and frequency and developing appropriate summary scores to best reflect these factors
based on reports from the Time Line Follow Back surveys. Latent class analyses (LCA) with
maximum likelihood estimation and robust standard errors will be used to explore
parent-student communication patterns using those summary scores. The investigators will
derive empirically based classes of participants based on content, emotional tone, timing,
duration, mode, and frequency. This will involve estimating a series of models to determine
the number of unique classes that best fit the data by comparing the Bayesian Information
Criterion, the Lo-Mendell-Rubin adjusted Likelihood Ratio test and the Parametric
Bootstrapped Likelihood Ratio test across models and selecting the most parsimonious model
with the best fit.

Inclusion Criteria:

- Students at least 17 and under 22 years of age who are attending college for first
time

Exclusion Criteria:

- Non-English speaking students/parents

- Students whose primary residence is outside the U.S.

- Students who will be living at home rather than at school